On Fri 19-02-21 12:17:11, Oscar Salvador wrote:

On Fri, Feb 19, 2021 at 11:55:00AM +0100, Michal Hocko wrote:

quoted

It is not the lock that I care about but more about counters. The
intention was that there is a single place to handle both enqueing and
dequeing. As not all places require counters to be updated. E.g. the
migration which just replaces one page by another.

I see.
alloc_fresh_huge_page->prep_new_huge_page increments h->nr_huge_pages{_node}
counters.
Which means:

quoted

      new_page = alloc_fresh_huge_page();
      if (!new_page)
              goto fail;
      spin_lock(hugetlb_lock);
      if (!PageHuge(old_page)) {
              /* freed from under us, nothing to do */ 
              __update_and_free_page(new_page);

Here we need update_and_free_page, otherwise we would be leaving a stale value
in h->nr_huge_pages{_node}. 

quoted

              goto unlock;
      }
      list_del(&old_page->lru);
      __update_and_free_page(old_page);

Same here.

quoted

      __enqueue_huge_page(new_page);

This is ok since h->free_huge_pages{_node} do not need to be updated.

Fair enough. I didn't get to think this through obviously, but you
should get the idea ;)
-- 
Michal Hocko
SUSE Labs